Pretrained model has fewer layers than actual model

[smit-r]

I trained the model on custom data and when I tested it, the results are not very good and it shows the following message: Pretrained network G has fewer layers; The following are not initialized: ['model1_1', 'model1_2']

[afaq-ahmad]

facing the same problem, did anybody know how to solve it

[smit-r]

Just ensure that you set --netG parameter same in training and testing scripts While training I had put --netG local and I used default value of --netG which was different

[smit-r]

If --netG is same, check that --ngf and --nef parameters are same while training and testing

[afaq-ahmad]

So i was using 2 n_local_enhancers after adding this option in test.py file like this solved the problem opt.n_local_enhancers=2

Thanks

[Akshace]

I fixed the same issue in mine by correcting the --ngf value in testing script. It should match with the one during training.